fix(custom_httpx): preserve non-ascii response headers

litellm/llms/custom_httpx/aiohttp_transport.py

@@ -341,7 +341,7 @@ class LiteLLMAiohttpTransport(AiohttpTransport):
 
         return httpx.Response(
             status_code=response.status,
-            headers=response.headers,
+            headers=response.raw_headers,
             stream=AiohttpResponseStream(response),
             request=request,
         )

litellm/llms/custom_httpx/http_handler.py

@@ -381,11 +381,11 @@ class MaskedHTTPStatusError(httpx.HTTPStatusError):
         except Exception:
             response_content = b""
 
-        response_headers = {
-            k: v
-            for k, v in original_error.response.headers.items()
-            if k.lower() not in ("content-encoding", "content-length")
-        }
+        response_headers = [
+            (k, v)
+            for k, v in original_error.response.headers.raw
+            if k.lower() not in (b"content-encoding", b"content-length")
+        ]
 
         masked_request = httpx.Request(
             method=original_error.request.method,

tests/test_litellm/llms/custom_httpx/test_aiohttp_transport.py

@@ -238,6 +238,7 @@ async def test_handle_async_request_uses_env_proxy(monkeypatch):
             class Resp:
                 status = 200
                 headers = {}
+                raw_headers = ()
 
                 async def __aenter__(self):
                     return self
@@ -262,6 +263,51 @@ async def test_handle_async_request_uses_env_proxy(monkeypatch):
     assert captured["proxy"] == proxy_url
 
 
+@pytest.mark.asyncio
+async def test_handle_async_request_preserves_non_ascii_response_headers():
+    """Aiohttp transport should pass raw response header bytes through to httpx."""
+    header_value = "本地化消息"
+
+    class FakeSession:
+        closed = False
+
+        def __init__(self):
+            try:
+                self._loop = asyncio.get_running_loop()
+            except RuntimeError:
+                self._loop = None
+
+        def request(self, *args, **kwargs):
+            class Resp:
+                status = 200
+                headers = {"X-Localized-Message": header_value}
+                raw_headers = ((b"X-Localized-Message", header_value.encode("utf-8")),)
+
+                async def __aenter__(self):
+                    return self
+
+                async def __aexit__(self, exc_type, exc, tb):
+                    pass
+
+                @property
+                def content(self):
+                    class C:
+                        async def iter_chunked(self, size):
+                            yield b"ok"
+
+                    return C()
+
+            return Resp()
+
+    transport = LiteLLMAiohttpTransport(client=lambda: FakeSession())  # type: ignore
+    response = await transport.handle_async_request(
+        httpx.Request("GET", "http://example.com/asset")
+    )
+
+    assert response.status_code == 200
+    assert response.headers.get("x-localized-message") == header_value
+
+
 @pytest.mark.asyncio
 async def test_handle_async_request_uses_env_proxy_per_url(monkeypatch):
     """Aiohttp transport should honor HTTP(S)_PROXY env vars unless NO_PROXY matches"""
@@ -295,6 +341,7 @@ async def test_handle_async_request_uses_env_proxy_per_url(monkeypatch):
             class Resp:
                 status = 200
                 headers = {}
+                raw_headers = ()
 
                 async def __aenter__(self):
                     return self
@@ -353,6 +400,7 @@ def _make_mock_response(should_fail=False, fail_count={"count": 0}):
     class MockResp:
         status = 200
         headers = {}
+        raw_headers = ()
 
         async def __aenter__(self):
             if should_fail and fail_count["count"] < 1:

tests/test_litellm/llms/custom_httpx/test_credential_leak_prevention.py

@@ -135,6 +135,24 @@ class TestMaskedHTTPStatusError:
         # Content-Encoding must have been stripped from the rebuilt headers.
         assert "content-encoding" not in {k.lower() for k in masked.response.headers}
 
+    def test_preserves_non_ascii_response_headers(self):
+        header_value = "本地化错误消息"
+        request = httpx.Request("GET", "https://api.example.com?key=SECRET")
+        response = httpx.Response(
+            status_code=400,
+            content=b"bad request",
+            headers=[
+                (b"x-localized-message", header_value.encode("utf-8")),
+            ],
+            request=request,
+        )
+        orig = httpx.HTTPStatusError("400", request=request, response=response)
+
+        masked = MaskedHTTPStatusError(orig)
+
+        assert masked.response.headers.get("x-localized-message") == header_value
+        assert "SECRET" not in str(masked.response.request.url)
+
 
 class TestSafeResponseHelpers:
     def test_safe_get_response_text_normal(self):

tests/test_litellm/test_streaming_connection_cleanup.py

@@ -40,6 +40,7 @@ async def test_aiohttp_transport_response_uses_stream_not_content():
             class Resp:
                 status = 200
                 headers = {}
+                raw_headers = ()
 
                 async def __aenter__(self):
                     return self
@@ -73,6 +74,7 @@ async def test_aiohttp_response_stream_aclose_releases_connection():
     class MockResponse:
         status = 200
         headers = {}
+        raw_headers = ()
 
         @property
         def content(self):